[PATCH 0/7] RFC Memory tagging support

[Richard Earnshaw]

On 16/06/2020 14:44, Florian Weimer via Libc-alpha wrote:
> * Szabolcs Nagy:
> 
>> The 06/15/2020 15:40, Richard Earnshaw wrote:
>>> 2) Tests that assume that malloc_usable_size will return a specific
>>> amount of free space.  The assumptions are not correct, because the
>>> tag colouring boundaries needed for MTE means that the 8 bytes in the
>>> block containing the back pointer can no-longer be used by users when
>>> we have MTE (they have a different colour that belongs to the malloc
>>> data structures).
>>
>> with --enable-memory-tagging i see
>>
>> FAIL: malloc/tst-malloc-usable
>> FAIL: malloc/tst-malloc-usable-static
>> FAIL: malloc/tst-malloc-usable-static-tunables
>> FAIL: malloc/tst-malloc-usable-tunables
>>
>> malloc_usable_size(malloc(7)) is 16 with
>> MALLOC_CHECK_=0 and it's 0 with MALLOC_CHECK_=3.
>>
>> i think this breaks existing usage, so either
>> malloc check should be disabled if memory tagging
>> is enabled or fixed to be compatible.
>> (or at least the issue should be documented)
> 
> I'm with Richard here—this is an incorrect test expectation, not a bug
> in the implementation.
> 
> Thanks,
> Florian
> 

Actually, I think there is a real issue that I have to solve: the usable
size should never be less than the allocation request.

The problem is that I round down the allocation agressively in
malloc_usable_size and that does not account for the MALLOC_CHECK case
where the overall size is reduced by one to allow for the magic cookie
marker.

When MTE is enabled, I'm not sure it makes too much sense to also enable
MALLOC_CHECK: it does essentially the same thing, but less well.  But
when it is off (and the library is configured to support MTE), it does
need to work as expected.

I'm still thinking about this case.  One option is to put the cookie
inside the word that we no-longer hand out to the user.  It's harmless
to put it there and it avoids wasting even more space, even if the user
normally can't overrun it when MTE is on.

R.